it does not say what types of cameras/lenses are required for the system. some lenses have distortions introduced into the images depending on the f-stop used during photography.
i wonder if these distortions will be taken of in the software. can someone using the demo confirm this?
thanks.
Photo-Based Point Clouds
😉
> it does not say what types of cameras/lenses are required for the system. some lenses have distortions introduced into the images depending on the f-stop used during photography.
There's a 40+ minute video on YouTube that walks through a demo. It was created not by Datumate but by a vendor, and in it they address the camera issue a bit. They recommend a high-pixel-count (20mp or so) camera with a fixed-focus lens. Their example is a Samsung model that sells for around $450.
They do have a calibration procedure for the particular camera that you are going to be using.
After watching a vendor video of the DatuGram 3D software, one thing that strikes me is that the level of effort required to accurately (as defined by the capability of the technology) locate any point in the photos is very high. Each discrete point has to be identified in multiple photos and the resulting solution statistics checked for acceptability.
This would be fine for situations in which the desired points aren't accessible to field instruments (e.g. reflectorless total station) or can't be identified at the time of the field work, but for most topo situations in which I'm involved, it looks like it'd be faster and more accurate to measure the points directly in the field.
It's interesting technology, but it's not going to replace field surveying or laser scanning. If there's a place for it in my toolbox, I haven't found it yet.
Jim,
You are exactly correct. In fact, the best results are achieved with "pre-targeted points" in which reflector tape is placed at all salient points. Each point must be imaged on at least three photos to have any sort of accuracy statistic computed, and best results are from points imaged on about 6 to 9 different images. For "topo" applications, it's mostly useless unless you can pick out discrete identifiable points such as rocks or distinctive pebbles. Photogrammetry works, but it's still dependent on geometry. Hardest thing to achieve is good results from non-targeted points. Pray for rivets or corners with sharp edges, and lots of them.
Isn't a lot of this pre-targeting being replaced with pixel pattern algorithms?
Very problematic with highly convergent imagery due to different angles of illumination. Tends to work well with vertical aerial photography, less so with terrestrial or low-altitude convergent photography, especially if there is significant depth in the subject volume as a proportion of the camera-to-object distance. Works nearly perfect with flat desert areas. 🙂
> Isn't a lot of this pre-targeting being replaced with pixel pattern algorithms?
as i recall, Datumate has computations for both relative pixel to coordinate transformations and absolute pixel to coordinate. with the proper combination of imagery, laps, and measurements, a model can be generated in absolute coordinate systems
I saw a demo of Datumate. My mind raced with the possibilities. Reading these posts makes me realize that the usefulness of this possible tool in one’s kit is the applications of the tool. We model a lot of intricate structures. Lattice electrical towers, monopole electric towers, substations and power plants. These features might be the ideal target for this application. I hope so. I am investigating the technology along with drone-based cameras to capture the images. Flying a quadcopter around high-voltage transmission lines might be a shocking experience. 😉
